ResearchBenchmarksSecure Code Generation on SecurityEval PythonFollow75.8SPR@1TSP54.16859.78465.471.016Jun 2, 2026Evaluation ResultsMethodMethodLinksSPR@1TSPModel Backbone=CodeLla...Model Backbone=CodeLlama-7B2026.0675.8SAFECODERModel Backbone=CodeLla...Model Backbone=CodeLlama-7B2026.0673.7SELF-PLAYModel Backbone=CodeLla...Model Backbone=CodeLlama-7B2026.0669.6SFTModel Backbone=CodeLla...Model Backbone=CodeLlama-7B2026.0657BASE LLMModel Backbone=CodeLla...Model Backbone=CodeLlama-7B2026.0655